Reduksi Noise Data Curah Hujan di Kota Bandung Menggunakan Variational Mode Decomposition (VMD)
DOI:
https://doi.org/10.29313/bcss.v6i2.25669Keywords:
Variational Mode Decomposition, curah hujan, dekomposisi sinyalAbstract
Abstract. Daily rainfall data in Bandung City exhibit nonlinear and nonstationary characteristics caused by seasonal patterns and extreme fluctuations, making accurate forecasting a challenging task. To address this issue, this study applies Variational Mode Decomposition (VMD) as a preprocessing technique to reduce data complexity and noise before forecasting. VMD decomposes the original rainfall signal into several Intrinsic Mode Functions (IMFs) through an adaptive variational optimization process, enabling a more stable decomposition while overcoming the mode-mixing problem commonly found in Empirical Mode Decomposition (EMD). The optimal number of decomposition modes (K) was determined using three quantitative evaluation criteria proposed by Wu et al. (2020), namely energy entropy, minimum gap center frequency, and minimum energy proportion, with K values ranging from 2 to 6. The results indicate that K = 2 is the optimal choice, producing the lowest energy entropy (0.2480), the largest minimum gap center frequency (0.16486), and no redundant IMF, as indicated by the minimum energy proportion of 6.78%, which exceeds the 1% threshold. The decomposition separates the rainfall signal into IMF1, which captures the dominant long-term trend, and IMF2, which represents high-frequency fluctuations associated with noise. These findings demonstrate that VMD effectively simplifies complex rainfall data into more homogeneous signal components, providing a more suitable representation for subsequent forecasting models.
Keywords: Variational Mode Decomposition, rainfall, signal decomposition
Abstrak. Data curah hujan harian di Kota Bandung memiliki karakteristik nonlinier dan non-stasioner akibat pengaruh variasi musiman serta fluktuasi yang tinggi, sehingga proses prediksi menjadi lebih kompleks. Untuk mengatasi permasalahan tersebut, penelitian ini menerapkan Variational Mode Decomposition (VMD) sebagai tahap preprocessing untuk mereduksi kompleksitas dan noise sebelum proses forecasting. Metode VMD mendekomposisi sinyal curah hujan menjadi beberapa Intrinsic Mode Function (IMF) melalui pendekatan optimasi variasional yang adaptif, sehingga mampu menghasilkan dekomposisi yang lebih stabil sekaligus mengurangi permasalahan mode mixing yang umum terjadi pada Empirical Mode Decomposition (EMD). Penentuan jumlah mode optimal (K) dilakukan menggunakan tiga kriteria evaluasi yang dijelaskan oleh Wu et al. (2020), yaitu energy entropy, minimum gap center frequency, dan proporsi energi minimum, dengan pengujian pada K = 2 hingga 6. Hasil penelitian menunjukkan bahwa K = 2 merupakan jumlah mode yang paling optimal karena menghasilkan energy entropy terkecil (0,2480), minimum gap center frequency terbesar (0,16486), serta tidak ditemukan IMF redundan dengan proporsi energi minimum sebesar 6,78% yang berada di atas ambang batas 1%. Hasil dekomposisi memisahkan sinyal curah hujan menjadi IMF1 yang merepresentasikan tren jangka panjang dan IMF2 yang menggambarkan fluktuasi berfrekuensi tinggi yang menyerupai noise. Temuan ini menunjukkan bahwa VMD mampu menyederhanakan data curah hujan yang kompleks menjadi komponen yang lebih homogen sehingga lebih sesuai digunakan sebagai masukan pada tahap pemodelan prediksi selanjutnya.
Kata Kunci: Variational Mode Decomposition, curah hujan, dekomposisi sinyal
References
BMKG. (2020). Data dan Informasi Curah Hujan. BMKG.
BNPB. (2025). Data Bencana Indonesia 2024.
Dragomiretskiy, K., & Zosso, D. (2014). Variational Mode Decomposition. IEEE Transactions on Signal Processing, 62(3), 531–544. https://doi.org/10.1109/TSP.2013.2288675
Hyndman, R. J., & Athanasopoulos, G. (2018). Forecasting: Principles and Practice (3rd ed.). OTexts.
Rakhim, R., Pattipeilohy, J., Perubahan, I., Musim, P., Distribusi, D. A. N., Agroteknologi, P. S., Gunung, J., Amban, S., Rakhim, R., & Pattipeilohy, J. (2022). Identification Of Changes In Seasonal Patterns And Frequency Distribution Of Rainfall In Manokwari. 3(5), 35–43.
Wani, O. A., Mahdi, S. S., Kumar, S. S., dkk. (2024). Predicting Rainfall Using Machine Learning, Deep Learning, and Time Series Models Across an Altitudinal Gradient in the North-Western Himalayas.
Wu, S., Feng, F., Zhu, J., Wu, C., & Zhang, G. (2020). A Method for Determining Intrinsic Mode Function Number in Variational Mode Decomposition and Its Application to Bearing Vibration Signal Processing. Shock and Vibration, 2020, Article 8304903. https://doi.org/10.1155/2020/8304903
Zhang, Y., Hu, K., Lu, L., & Yang, Q. (2025). Seasonally Adaptive VMD-SSA-LSTM : A Hybrid Deep Learning Framework for High-Accuracy District Heating Load Forecasting. https://doi.org/10.3390/math13152406